handler.py

from typing import Dict, List, Any
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, StoppingCriteria, StoppingCriteriaList


class EndpointHandler():
    def __init__(self, path=""):
        tokenizer = AutoTokenizer.from_pretrained(path)
        tokenizer.pad_token = tokenizer.eos_token
        self.model = AutoModelForCausalLM.from_pretrained(path).to('cuda')
        self.tokenizer = tokenizer
        self.stopping_criteria = StoppingCriteriaList([StopAtPeriodCriteria(tokenizer)])

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            inputs (:obj: `str`)
            kwargs
      Return:
            A :obj:`list` | `dict`: will be serialized and returned
        """
        inputs = data.pop("inputs", data)
        additional_bad_words_ids = data.pop("additional_bad_words_ids", [])


        # 3070, 10456, [313, 334] corresponds to "(*", and we do not want to output a comment
        # 13 is a newline character
        # [1976, 441, 29889], [4920, 441, 29889] is "Abort." [4920, 18054, 29889] is "Aborted."
        # [2087, 29885, 4430, 29889] is "Admitted."
        bad_words_ids = [[3070], [313, 334], [10456], [13], [1976, 441, 29889], [2087, 29885, 4430, 29889], [4920, 441], [4920, 441, 29889], [4920, 18054, 29889]]
        bad_words_ids.extend(additional_bad_words_ids)

        input_ids = self.tokenizer.encode(inputs, return_tensors="pt").to('cuda')
        max_generation_length = 75  # Desired number of tokens to generate
        # max_input_length = 4092 - max_generation_length  # Maximum input length to allow space for generation

        # # Truncate input_ids to the most recent tokens that fit within the max_input_length
        # if input_ids.shape[1] > max_input_length:
        #     input_ids = input_ids[:, -max_input_length:]

        max_length = input_ids.shape[1] + max_generation_length
        
        generated_ids = self.model.generate(
            input_ids,
            max_length=max_length,  # 50 new tokens
            bad_words_ids=bad_words_ids,
            temperature=1,
            top_k=40,
            do_sample=True,
            stopping_criteria=self.stopping_criteria,
        )

        generated_text = self.tokenizer.decode(generated_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
        prediction = [{"generated_text": generated_text, "generated_ids": generated_ids[0][input_ids.shape[1]:].tolist()}]
        return prediction


class StopAtPeriodCriteria(StoppingCriteria):
    def __init__(self, tokenizer):
        self.tokenizer = tokenizer

    def __call__(self, input_ids, scores, **kwargs):
        # Decode the last generated token to text
        last_token_text = self.tokenizer.decode(input_ids[:, -1], skip_special_tokens=True)
        # Check if the decoded text ends with a period
        return '.' in last_token_text